GPU use and billing understanding

I have some billing doubts about deploying models on paperspace. Referring to Deploying Deep Learning Models on the Web With Flask | Paperspace Blog and Deploying Deep Learning Models Part 2: Hosting on Paperspace
I would like to know, if you deploy a flask app that makes some machine learning calculations when the user calls for an action. Do you pay for every hour of the day and every day of the month as the app is always available and listening for requests? Or do you pay only for the use of the GPU, that is in the example, only when the user draws a digit until the model gets its result?

I’m developing an Android app where the user can select an image that has to be sent, processed by a pytorch model and send back the result. The server has to be always available but I only want to pay for the actual use of GPU.

I know this question is pretty basic and vague but I feel a little bit lost about deploying this kind of machine learning backends.